Binary Safe/Unsafe Classification on R-Judge (test)
57.8AccuracyBraveGuard-Qwen3-Guard-8B
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| BraveGuard-Qwen3-Guard-8BBackbone=Qwen3-Guard, Parameters=8B2026.05 | 57.8 | 91.2 | 69.7 | |
| NemoGuard2026.05 | 54.4 | 40.6 | 48.5 | |
| Llama3.1-8B-InstructParameters=8B2026.05 | 53.7 | 100 | 69.5 | |
| Qwen3-Guard2026.05 | 40.6 | 5.5 | 9 |